internalTrafficPolicy: Local - time for error

General Discussions
,
1

Cluster information:

Kubernetes version: 1.2.6
Cloud being used: eks
Hello everyone,
I’m trying to test internalTrafficPolicy: Local and see what happens when there are available pods on the specific node:I exec into a pod on a specific node and ran the curl command:
curl micronaut-k8-sample:8080/stickiness/5 and got:
curl: (28) Failed to connect to micronaut-k8-sample port 8080 after 129261 ms: Couldn’t connect to server
(This api succeeds when running on a node that there are available pods of for service micronaut-k8-sample)
Is there a way to configure kube poxy to get the error faster? Waiting more than 2 minutes is a bit of an issue, but I also have some other http calls for that service that might need waiting when service is available, so don’t want to just define a timeout

2

I think it is currently the policy that, in this case requests to a service with no local endpoints are dropped rather than ICMP rejected. I’m not sure why, frankly.

You should file a bug against GitHub - kubernetes/kubernetes: Production-Grade Container Scheduling and Management and we can discuss there.

3

Thanks. Raised: Requests to a service with no local endpoints are dropped rather than ICMP rejected · Issue #121371 · kubernetes/kubernetes · GitHub